This paper was accepted for publication by the Journal on Systems and Software on 18 November 2024 and is already available online.

Abstract. Identifying whether GitHub contributors are automated bots is important for empirical research on collaborative software development practices. Multiple such bot identification approaches have been proposed in the past. In this article, we identify the limitations of these approaches and we propose a new binary classification model, called BIMBAS, to overcome these limitations. To do so, we propose a new ground-truth dataset containing 1,035 bots and 1,115 humans on GitHub. We train BIMBAS on a wide range of features extracted from the activity sequences of these GitHub contributors. We show that the performance of BIMBAS (in terms of precision, recall, F1 score and AUC) is comparable to state-of-the-art bot identification approaches, while being able to identify bots engaged in a wider range of activity types. We implement RABBIT, an open-source command-line bot identification tool based on BIMBAS. We demonstrate its ability to be used at scale, and show that its efficiency outperforms the state-of-the-art.